1 /* mdesc.c: Sun4V machine description handling.
3 * Copyright (C) 2007 David S. Miller <davem@davemloft.net>
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/bootmem.h>
8 #include <linux/log2.h>
9 #include <linux/list.h>
10 #include <linux/slab.h>
13 #include <asm/hypervisor.h>
14 #include <asm/mdesc.h>
16 #include <asm/oplib.h>
19 /* Unlike the OBP device tree, the machine description is a full-on
20 * DAG. An arbitrary number of ARCs are possible from one
21 * node to other nodes and thus we can't use the OBP device_node
22 * data structure to represent these nodes inside of the kernel.
24 * Actually, it isn't even a DAG, because there are back pointers
25 * which create cycles in the graph.
27 * mdesc_hdr and mdesc_elem describe the layout of the data structure
28 * we get from the Hypervisor.
31 u32 version; /* Transport version */
32 u32 node_sz; /* node block size */
33 u32 name_sz; /* name block size */
34 u32 data_sz; /* data block size */
35 } __attribute__((aligned(16)));
39 #define MD_LIST_END 0x00
41 #define MD_NODE_END 0x45
43 #define MD_PROP_ARC 0x61
44 #define MD_PROP_VAL 0x76
45 #define MD_PROP_STR 0x73
46 #define MD_PROP_DATA 0x64
59 struct mdesc_mem_ops {
60 struct mdesc_handle *(*alloc)(unsigned int mdesc_size);
61 void (*free)(struct mdesc_handle *handle);
65 struct list_head list;
66 struct mdesc_mem_ops *mops;
69 unsigned int handle_size;
70 struct mdesc_hdr mdesc;
73 static void mdesc_handle_init(struct mdesc_handle *hp,
74 unsigned int handle_size,
77 BUG_ON(((unsigned long)&hp->mdesc) & (16UL - 1));
79 memset(hp, 0, handle_size);
80 INIT_LIST_HEAD(&hp->list);
82 atomic_set(&hp->refcnt, 1);
83 hp->handle_size = handle_size;
86 static struct mdesc_handle *mdesc_bootmem_alloc(unsigned int mdesc_size)
88 struct mdesc_handle *hp;
89 unsigned int handle_size, alloc_size;
91 handle_size = (sizeof(struct mdesc_handle) -
92 sizeof(struct mdesc_hdr) +
94 alloc_size = PAGE_ALIGN(handle_size);
96 hp = __alloc_bootmem(alloc_size, PAGE_SIZE, 0UL);
98 mdesc_handle_init(hp, handle_size, hp);
103 static void mdesc_bootmem_free(struct mdesc_handle *hp)
105 unsigned int alloc_size, handle_size = hp->handle_size;
106 unsigned long start, end;
108 BUG_ON(atomic_read(&hp->refcnt) != 0);
109 BUG_ON(!list_empty(&hp->list));
111 alloc_size = PAGE_ALIGN(handle_size);
113 start = (unsigned long) hp;
114 end = start + alloc_size;
116 while (start < end) {
119 p = virt_to_page(start);
120 ClearPageReserved(p);
126 static struct mdesc_mem_ops bootmem_mdesc_memops = {
127 .alloc = mdesc_bootmem_alloc,
128 .free = mdesc_bootmem_free,
131 static struct mdesc_handle *mdesc_kmalloc(unsigned int mdesc_size)
133 unsigned int handle_size;
136 handle_size = (sizeof(struct mdesc_handle) -
137 sizeof(struct mdesc_hdr) +
140 base = kmalloc(handle_size + 15, GFP_KERNEL);
142 struct mdesc_handle *hp;
145 addr = (unsigned long)base;
146 addr = (addr + 15UL) & ~15UL;
147 hp = (struct mdesc_handle *) addr;
149 mdesc_handle_init(hp, handle_size, base);
156 static void mdesc_kfree(struct mdesc_handle *hp)
158 BUG_ON(atomic_read(&hp->refcnt) != 0);
159 BUG_ON(!list_empty(&hp->list));
161 kfree(hp->self_base);
164 static struct mdesc_mem_ops kmalloc_mdesc_memops = {
165 .alloc = mdesc_kmalloc,
169 static struct mdesc_handle *mdesc_alloc(unsigned int mdesc_size,
170 struct mdesc_mem_ops *mops)
172 struct mdesc_handle *hp = mops->alloc(mdesc_size);
180 static void mdesc_free(struct mdesc_handle *hp)
185 static struct mdesc_handle *cur_mdesc;
186 static LIST_HEAD(mdesc_zombie_list);
187 static DEFINE_SPINLOCK(mdesc_lock);
189 struct mdesc_handle *mdesc_grab(void)
191 struct mdesc_handle *hp;
194 spin_lock_irqsave(&mdesc_lock, flags);
197 atomic_inc(&hp->refcnt);
198 spin_unlock_irqrestore(&mdesc_lock, flags);
202 EXPORT_SYMBOL(mdesc_grab);
204 void mdesc_release(struct mdesc_handle *hp)
208 spin_lock_irqsave(&mdesc_lock, flags);
209 if (atomic_dec_and_test(&hp->refcnt)) {
210 list_del_init(&hp->list);
213 spin_unlock_irqrestore(&mdesc_lock, flags);
215 EXPORT_SYMBOL(mdesc_release);
217 void mdesc_update(void)
219 unsigned long len, real_len, status;
220 struct mdesc_handle *hp, *orig_hp;
223 (void) sun4v_mach_desc(0UL, 0UL, &len);
225 hp = mdesc_alloc(len, &kmalloc_mdesc_memops);
227 printk(KERN_ERR "MD: mdesc alloc fails\n");
231 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
232 if (status != HV_EOK || real_len > len) {
233 printk(KERN_ERR "MD: mdesc reread fails with %lu\n",
235 atomic_dec(&hp->refcnt);
240 spin_lock_irqsave(&mdesc_lock, flags);
244 if (atomic_dec_and_test(&orig_hp->refcnt))
247 list_add(&orig_hp->list, &mdesc_zombie_list);
248 spin_unlock_irqrestore(&mdesc_lock, flags);
251 static struct mdesc_elem *node_block(struct mdesc_hdr *mdesc)
253 return (struct mdesc_elem *) (mdesc + 1);
256 static void *name_block(struct mdesc_hdr *mdesc)
258 return ((void *) node_block(mdesc)) + mdesc->node_sz;
261 static void *data_block(struct mdesc_hdr *mdesc)
263 return ((void *) name_block(mdesc)) + mdesc->name_sz;
266 u64 mdesc_node_by_name(struct mdesc_handle *hp,
267 u64 from_node, const char *name)
269 struct mdesc_elem *ep = node_block(&hp->mdesc);
270 const char *names = name_block(&hp->mdesc);
271 u64 last_node = hp->mdesc.node_sz / 16;
274 if (from_node == MDESC_NODE_NULL) {
276 } else if (from_node >= last_node) {
277 return MDESC_NODE_NULL;
279 ret = ep[from_node].d.val;
282 while (ret < last_node) {
283 if (ep[ret].tag != MD_NODE)
284 return MDESC_NODE_NULL;
285 if (!strcmp(names + ep[ret].name_offset, name))
289 if (ret >= last_node)
290 ret = MDESC_NODE_NULL;
293 EXPORT_SYMBOL(mdesc_node_by_name);
295 const void *mdesc_get_property(struct mdesc_handle *hp, u64 node,
296 const char *name, int *lenp)
298 const char *names = name_block(&hp->mdesc);
299 u64 last_node = hp->mdesc.node_sz / 16;
300 void *data = data_block(&hp->mdesc);
301 struct mdesc_elem *ep;
303 if (node == MDESC_NODE_NULL || node >= last_node)
306 ep = node_block(&hp->mdesc) + node;
308 for (; ep->tag != MD_NODE_END; ep++) {
320 val = data + ep->d.data.data_offset;
321 len = ep->d.data.data_len;
330 if (!strcmp(names + ep->name_offset, name)) {
339 EXPORT_SYMBOL(mdesc_get_property);
341 u64 mdesc_next_arc(struct mdesc_handle *hp, u64 from, const char *arc_type)
343 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
344 const char *names = name_block(&hp->mdesc);
345 u64 last_node = hp->mdesc.node_sz / 16;
347 if (from == MDESC_NODE_NULL || from >= last_node)
348 return MDESC_NODE_NULL;
353 for (; ep->tag != MD_NODE_END; ep++) {
354 if (ep->tag != MD_PROP_ARC)
357 if (strcmp(names + ep->name_offset, arc_type))
363 return MDESC_NODE_NULL;
365 EXPORT_SYMBOL(mdesc_next_arc);
367 u64 mdesc_arc_target(struct mdesc_handle *hp, u64 arc)
369 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
375 EXPORT_SYMBOL(mdesc_arc_target);
377 const char *mdesc_node_name(struct mdesc_handle *hp, u64 node)
379 struct mdesc_elem *ep, *base = node_block(&hp->mdesc);
380 const char *names = name_block(&hp->mdesc);
381 u64 last_node = hp->mdesc.node_sz / 16;
383 if (node == MDESC_NODE_NULL || node >= last_node)
387 if (ep->tag != MD_NODE)
390 return names + ep->name_offset;
392 EXPORT_SYMBOL(mdesc_node_name);
394 static void __init report_platform_properties(void)
396 struct mdesc_handle *hp = mdesc_grab();
397 u64 pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform");
401 if (pn == MDESC_NODE_NULL) {
402 prom_printf("No platform node in machine-description.\n");
406 s = mdesc_get_property(hp, pn, "banner-name", NULL);
407 printk("PLATFORM: banner-name [%s]\n", s);
408 s = mdesc_get_property(hp, pn, "name", NULL);
409 printk("PLATFORM: name [%s]\n", s);
411 v = mdesc_get_property(hp, pn, "hostid", NULL);
413 printk("PLATFORM: hostid [%08lx]\n", *v);
414 v = mdesc_get_property(hp, pn, "serial#", NULL);
416 printk("PLATFORM: serial# [%08lx]\n", *v);
417 v = mdesc_get_property(hp, pn, "stick-frequency", NULL);
418 printk("PLATFORM: stick-frequency [%08lx]\n", *v);
419 v = mdesc_get_property(hp, pn, "mac-address", NULL);
421 printk("PLATFORM: mac-address [%lx]\n", *v);
422 v = mdesc_get_property(hp, pn, "watchdog-resolution", NULL);
424 printk("PLATFORM: watchdog-resolution [%lu ms]\n", *v);
425 v = mdesc_get_property(hp, pn, "watchdog-max-timeout", NULL);
427 printk("PLATFORM: watchdog-max-timeout [%lu ms]\n", *v);
428 v = mdesc_get_property(hp, pn, "max-cpus", NULL);
430 printk("PLATFORM: max-cpus [%lu]\n", *v);
438 if (max_cpu > NR_CPUS)
443 for (i = 0; i < max_cpu; i++)
444 cpu_set(i, cpu_possible_map);
451 static int inline find_in_proplist(const char *list, const char *match, int len)
456 if (!strcmp(list, match))
458 l = strlen(list) + 1;
465 static void __devinit fill_in_one_cache(cpuinfo_sparc *c,
466 struct mdesc_handle *hp,
469 const u64 *level = mdesc_get_property(hp, mp, "level", NULL);
470 const u64 *size = mdesc_get_property(hp, mp, "size", NULL);
471 const u64 *line_size = mdesc_get_property(hp, mp, "line-size", NULL);
475 type = mdesc_get_property(hp, mp, "type", &type_len);
479 if (find_in_proplist(type, "instn", type_len)) {
480 c->icache_size = *size;
481 c->icache_line_size = *line_size;
482 } else if (find_in_proplist(type, "data", type_len)) {
483 c->dcache_size = *size;
484 c->dcache_line_size = *line_size;
489 c->ecache_size = *size;
490 c->ecache_line_size = *line_size;
500 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
501 u64 target = mdesc_arc_target(hp, a);
502 const char *name = mdesc_node_name(hp, target);
504 if (!strcmp(name, "cache"))
505 fill_in_one_cache(c, hp, target);
510 static void __devinit mark_core_ids(struct mdesc_handle *hp, u64 mp,
515 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
516 u64 t = mdesc_arc_target(hp, a);
520 name = mdesc_node_name(hp, t);
521 if (!strcmp(name, "cpu")) {
522 id = mdesc_get_property(hp, t, "id", NULL);
524 cpu_data(*id).core_id = core_id;
528 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_BACK) {
529 u64 n = mdesc_arc_target(hp, j);
532 n_name = mdesc_node_name(hp, n);
533 if (strcmp(n_name, "cpu"))
536 id = mdesc_get_property(hp, n, "id", NULL);
538 cpu_data(*id).core_id = core_id;
544 static void __devinit set_core_ids(struct mdesc_handle *hp)
550 mdesc_for_each_node_by_name(hp, mp, "cache") {
555 level = mdesc_get_property(hp, mp, "level", NULL);
559 type = mdesc_get_property(hp, mp, "type", &len);
560 if (!find_in_proplist(type, "instn", len))
563 mark_core_ids(hp, mp, idx);
569 static void __devinit mark_proc_ids(struct mdesc_handle *hp, u64 mp,
574 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
575 u64 t = mdesc_arc_target(hp, a);
579 name = mdesc_node_name(hp, t);
580 if (strcmp(name, "cpu"))
583 id = mdesc_get_property(hp, t, "id", NULL);
585 cpu_data(*id).proc_id = proc_id;
589 static void __devinit __set_proc_ids(struct mdesc_handle *hp,
590 const char *exec_unit_name)
596 mdesc_for_each_node_by_name(hp, mp, exec_unit_name) {
600 type = mdesc_get_property(hp, mp, "type", &len);
601 if (!find_in_proplist(type, "int", len) &&
602 !find_in_proplist(type, "integer", len))
605 mark_proc_ids(hp, mp, idx);
611 static void __devinit set_proc_ids(struct mdesc_handle *hp)
613 __set_proc_ids(hp, "exec_unit");
614 __set_proc_ids(hp, "exec-unit");
617 static void __devinit get_one_mondo_bits(const u64 *p, unsigned int *mask,
626 if (!val || val >= 64)
629 *mask = ((1U << val) * 64U) - 1U;
633 *mask = ((1U << def) * 64U) - 1U;
636 static void __devinit get_mondo_data(struct mdesc_handle *hp, u64 mp,
637 struct trap_per_cpu *tb)
641 val = mdesc_get_property(hp, mp, "q-cpu-mondo-#bits", NULL);
642 get_one_mondo_bits(val, &tb->cpu_mondo_qmask, 7);
644 val = mdesc_get_property(hp, mp, "q-dev-mondo-#bits", NULL);
645 get_one_mondo_bits(val, &tb->dev_mondo_qmask, 7);
647 val = mdesc_get_property(hp, mp, "q-resumable-#bits", NULL);
648 get_one_mondo_bits(val, &tb->resum_qmask, 6);
650 val = mdesc_get_property(hp, mp, "q-nonresumable-#bits", NULL);
651 get_one_mondo_bits(val, &tb->nonresum_qmask, 2);
654 void __devinit mdesc_fill_in_cpu_data(cpumask_t mask)
656 struct mdesc_handle *hp = mdesc_grab();
660 mdesc_for_each_node_by_name(hp, mp, "cpu") {
661 const u64 *id = mdesc_get_property(hp, mp, "id", NULL);
662 const u64 *cfreq = mdesc_get_property(hp, mp, "clock-frequency", NULL);
663 struct trap_per_cpu *tb;
673 if (cpuid >= NR_CPUS)
675 if (!cpu_isset(cpuid, mask))
678 /* On uniprocessor we only want the values for the
679 * real physical cpu the kernel booted onto, however
680 * cpu_data() only has one entry at index 0.
682 if (cpuid != real_hard_smp_processor_id())
687 c = &cpu_data(cpuid);
688 c->clock_tick = *cfreq;
690 tb = &trap_block[cpuid];
691 get_mondo_data(hp, mp, tb);
693 mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
694 u64 j, t = mdesc_arc_target(hp, a);
697 t_name = mdesc_node_name(hp, t);
698 if (!strcmp(t_name, "cache")) {
699 fill_in_one_cache(c, hp, t);
703 mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_FWD) {
704 u64 n = mdesc_arc_target(hp, j);
707 n_name = mdesc_node_name(hp, n);
708 if (!strcmp(n_name, "cache"))
709 fill_in_one_cache(c, hp, n);
714 cpu_set(cpuid, cpu_present_map);
722 sparc64_multi_core = 1;
728 smp_fill_in_sib_core_maps();
733 void __init sun4v_mdesc_init(void)
735 struct mdesc_handle *hp;
736 unsigned long len, real_len, status;
739 (void) sun4v_mach_desc(0UL, 0UL, &len);
741 printk("MDESC: Size is %lu bytes.\n", len);
743 hp = mdesc_alloc(len, &bootmem_mdesc_memops);
745 prom_printf("MDESC: alloc of %lu bytes failed.\n", len);
749 status = sun4v_mach_desc(__pa(&hp->mdesc), len, &real_len);
750 if (status != HV_EOK || real_len > len) {
751 prom_printf("sun4v_mach_desc fails, err(%lu), "
752 "len(%lu), real_len(%lu)\n",
753 status, len, real_len);
760 report_platform_properties();
763 mdesc_fill_in_cpu_data(mask);